Page 1 of 9

British Journal of Healthcare and Medical Research - Vol. 9, No. 5

Publication Date: October, 25, 2022

DOI:10.14738/jbemi.95.13309. Mahwish, N., Saherawala, B. A., & Jhancy, M. (2022). Clinical Decision Making in Dysmorphology- Emerging Role of Artificial

Intelligence. British Journal of Healthcare and Medical Research, 9(5). 366-374.

Services for Science and Education – United Kingdom

Clinical Decision Making in Dysmorphology- Emerging Role of

Artificial Intelligence

Nayesha Mahwish

Department of Pediatrics, Ras Al Khaimah College of Medical

Sciences (RAKCOMS), RAK Medical and Health Sciences

University (RAKMHSU), Ras Al Khaimah, United Arab Emirates

Batul Abdeali Saherawala

Department of Pediatrics, Ras Al Khaimah College of Medical

Sciences (RAKCOMS), RAK Medical and Health Sciences

University (RAKMHSU), Ras Al Khaimah, United Arab Emirates

Malay Jhancy

Department of Pediatrics, Ras Al Khaimah College of Medical

Sciences (RAKCOMS), RAK Medical and Health Sciences

University (RAKMHSU), Ras Al Khaimah, United Arab Emirates

ABSTRACT

The human genome codes for more than 22,000 genes, many of which have been

implicated in human diseases. These genetic diseases are often associated with

dysmorphic facial features. Dysmorphic features occur due to premature closure of

cranial sutures resulting in changes in skull shape and facial characteristics.

Assessment of dysmorphic features is a crucial component of genetic consultations.

This requires a great deal of clinical experience and expertise and tends to be

subjective. Artificial intelligence-based analysis can come in handy for quick and

accurate identification of dysmorphic features. This review explores the role played

by artificial intelligence in identifying dysmorphic facies and diagnosing various

genetic diseases in children.

Keywords: artificial intelligence, dysmorphism, facial recognition technology, genetic

disorders

INTRODUCTION

Genetic disorders affect a large proportion of the human population. Individuals affected by

these diseases suffer from multiple comorbidities such as congenital heart diseases, respiratory

problems, and developmental delays. Early diagnosis can help prevent these comorbidities

thus, improving the quality of life of these patients [1].

Dysmorphic facial features occur in over 1500 different human genetic syndromes. These

dysmorphic features are quite distinct to each disorder [2]. Downs syndrome, for example, is

distinguished by a flattened facial profile, upward slanting palpebral fissures, small ears, a

protruding tongue, and extremity variations [3]. Another genetic disorder, Noonan syndrome,

Page 2 of 9

367

Mahwish, N., Saherawala, B. A., & Jhancy, M. (2022). Clinical Decision Making in Dysmorphology- Emerging Role of Artificial Intelligence. British

Journal of Healthcare and Medical Research, 9(5). 366-374.

URL: http://dx.doi.org/10.14738/jbemi.95.13309

displays characteristics such as hypertelorism, down slanted palpebral fissures, ptosis, a

depressed and wide nasal bridge, and low set ears [4].

Identifying dysmorphic facial features aids in the early identification and diagnosis of genetic

disorders. It forms an essential component of genetic consultations. However, it requires a great

deal of clinical experience and expertise. There may be subjective variations in assessment of

dysmorphic facies by different clinicians [5]. Hence, recognizing dysmorphic features is a

daunting task.

With recent advances in the field of artificial intelligence, facial recognition systems have been

developed that assist in the screening and diagnosis of genetic disorders [6]. Deep learning

systems can identify and distinguish genetic conditions based solely on facial phenotyping.

They have been shown to be more accurate than human experts in phenotype recognition of

genetic diseases [7].

In this review, we summarize the role played by artificial intelligence (AI) in identification of

dysmorphic facial features in children with various genetic disorders.

METHODOLOGY & DATA ANALYSIS

Journal articles published from January 2010 to October 2021 about application of artificial

intelligence in identification of dysmorphism were collected from the databases ProQuest,

PubMed, Scopus and Medline using the key words Artificial intelligence OR AI AND

identification OR diagnosis AND dysmorphology OR dysmorphism AND children AND genetic

disorders. The articles were reviewed. The method is represented below in figure I.

Fig.I Methodology

RESULTS

Facial recognition technology

Artificial intelligence (AI) allows computers to solve complex problems and turn raw data into

meaningful data to be used to classify syndrome types. It uses convolutional neural networks

(CNNs) that have massive neurons, layers, and connectivity [8]. Systems such as DeepFace use

Keyword search (Artificial intelligence OR AI AND

identification OR diagnosis AND dysmorphology OR

dysmorphism AND children AND genetic disorders)

37 articles

4 articles

duplicated 33 articles

10 articles

excluded 23 articles

Page 3 of 9

368

British Journal of Healthcare and Medical Research (BJHMR) Vol 9, Issue 5, October - 2022

Services for Science and Education – United Kingdom

CNNs while being trained on large amounts of training data to reach performance at the level

of humans.

The most widely used facial recognition application by geneticists is Face2Gene [created by

Facial Dysmorphology Novel Analysis (FDNA Inc., Boston, Massachusetts, USA)] [9]. The

Face2Gene application is readily available to download onto a mobile device and can be used

worldwide wherever there is access to the internet. Face2Gene has a large and an ever-growing

of database of digital images of syndromic patients acting as a reference for the system. Real

life images are taken and processed by a landmark detection algorithm to geometrically

standardize the patient’s face for face verification and reduce pose variation. Ratios between

the points are calculated and compared to a holistic patient ratio. The system is then trained on

these images from a large data set with the help of clinicians identifying patients. The system

then can give a diagnosis on what a patient’s gestalt is best matched to a syndrome [10].

Genetic disorders identified by facial recognition technology

Artificial intelligence-based systems have been tried and tested for the identification of various

genetic disorders ranging from common disorders such as Turners syndrome to rare disorders

such as Phosphomannomutase-2 deficiency (Table I).The clinical trials have been discussed

below.

Noonan syndrome

Noonan syndrome is an autosomal dominant/recessive disorder. Some of its features include

short stature, craniofacial dysmorphism, cardiac abnormalities, short and/or webbed neck, and

cryptorchidism in male patients. This syndrome occurs due to mutations in genes encoding

proteins of the RAS-MAPK signaling pathway, resulting in pathway dysregulation. 15 such

genes have been identified so far: PTPN11, SOS1, RAF1, BRAF, HRAS, KRAS, NRAS, SHOC2,

MAP2K1, MAP2K2, CBL, RIT1, RASA2, A2ML1, and LZTR1.

In one study, facial images of 60 molecularly confirmed Chinese NS were evaluated with the

Face2Gene Research Application (FDNA Inc., Boston, Massachusetts). The images comprised

six pathogenic variants (PTPN11, SOS1, SHOC2, KRAS, RAF1, and RIT1) of the disorder. Results

showed that the mean accuracy that was achieved by Face2Gene on the original sample set was

28%. Moreover, each gene was accompanied with different facial features, all of which were

distinguished by the application. Patients with SHOC2 pathogenic variants were characterized

by significant macrocephaly and thin sparse hair, while patients with RAF1 pathogenic variants

had prominent foreheads [11].

Angelman syndrome

Angelman syndrome is a neurogenetic disorder characterized by developmental delay,

excessive laughter, absent or severely limited speech, seizures with a characteristic

electroencephalogram (EEG) and microcephaly. Characteristic facial features include wide

mouth, protruding tongue, midface recession and prognathism. AS occurs due to deficient

expression of gene UBE3A on chromosome 15. Different molecular subtypes of AS are

characterized by minor differences in facial features. Identification of these subtle facial

features could guide genetic testing for AS patients and their families.